Model S

The Model S will be able to Auto-Steer, but will require more sensors for Semi-Autonomous driving

edited March 2015 in Model S
As an engineer, I'm skeptical that the Model S, in it's present hardware configuration, can have semi-autonomous driving by this summer. There are simply not enough sensors in the present build for the kind of situational awareness required for it. Semi-autonomous driving is provided as a result of the sum of the data fed back from the sensors, and the collective data are analyzed thousands of times per second and tied back to the power drive unit, steering, friction braking, regen braking, and traction and stability control. The current vintage of Model S autopilot cars (Q4 2014 to Q1 2015) have one front radar, one front monocular Mobileye optical camera, and a 360-degree set of ultrasonic sensors. This present configuration works great for TACC and collision warning, but I believe still has many sensor blind spots for total sensing and active maneuvering required of autonomous driving. (On the other hand, I want to make the distinction that it would be able to Auto-steer.)

Mercedes-Benz have been at it since 1998, so I think their autopilot technology is the standard by which Tesla can be measured by. Here's how the Model S stacks up (below). Audi, Infiniti, Toyota are also realizing that these are the minimum set of eyes for semiautonomous driving. I hope my current Model S is future-proof to accommodate these systems as upgrades, much like how Mobileye can be installed and programmed off-the-shelf.

I'm also optimistic that Tesla can quickly pivot their builds and are presently working on this hardware setup. Tesla's prime advantage over Mercedes-Benz and other ICE car companies is that electric motors have much finer responses over combustion engines for situations requiring very precise wheel response. The characteristic lag response in gas engines can add precious milliseconds when time matters. Tesla can also update software at will. Digital > Analog.

I would like to hear the point of view of 2014 S550 owners that are using this technology.

image

The illustrations below are for the Mercedes Benz system.
(click for larger images)

image

image

image

More:
http://spectrum.ieee.org/transportation/self-driving/how-we-gave-sight-to-the-mercedes-robotic-car
http://spectrum.ieee.org/transportation/self-driving/how-selfdriving-cars-will-sneak-onto-our-roads
http://spectrum.ieee.org/transportation/self-driving/2014-top-ten-tech-cars
http://www.extremetech.com/extreme/161832-2014-mercedes-s-class-review-the-best-most-technologically-advanced-car-you-will-ever-drive


keywords: TACC, autopilot, autonomous driving, semiautonomous driving, autosteer, auto-steer, autonomous, semi-autonomous
---------------------------------------------------------
Evino Veritas.


UPDATE: Elon did not say autonomous or semi-autonomous driving by this summer. He said Auto-steer. Clear distinction there technically. Inaccurately, the media is parsing semantics again for sensational headlines, or they just don't understand.
«134

Comments

  • edited November -1
    i actually agree.

    There are a lot of situations in which Tesla's current sensor array is not enough. Snow, ice, heavy rain. Etc.

    You need multiple sources of data on the environment that the central car computer can utilize to create a dynamic assessment of the driving situation.

    But I think Tesla's headed in the right direction.
  • edited November -1
    I agree completely. The fact that the car can't see behind (except 5m sensors) or the sides, is a problem. I don't really mind if Tesla can't do this or that. That is not an issue that risks lives. The big issue is, Tesla is advertising things that it can't do, such as "automatically changing lanes". This is causing misunderstandings and I have proof of misunderstandings. Recently I sent them the following email. I think more people should contact them about this issue. If you scroll down below to the end of this page, there is a contact link.

    To Tesla Motors, Autonomous Driving Engineering Team,

    The Model S can not automatically change lanes. The driver is still required to check mirrors before initiating a lane change. However on Tesla website on the Model S page, in Tesla Blog posts and in Elon's interviews there is misleading information saying that Model S can change lanes automatically.

    [Post script addition: This gives the impression that checking mirrors is not needed. At least one person here in the forum misunderstood the issue and thought I was complaining that Tesla says the car will initiate lane change on its own but it can't. No that's not the problem. It is clear that the car will not initiate the lane change. The driver will. What is not clear is, the advertising material gives the impression there is no need to check the mirrors. Therefore some drivers might not check the mirrors before they perform an autopilot lane change using only the turn signal.]

    Media is repeating this incorrect information. This will cause accidents when this feature is enabled before people will assume they don't have to check mirrors.

    Here is a recent example of media misunderstanding, 17 Mar 2015:
    CNET: At 2:12 "Highway lane changing is interesting. You can do it by just signalling."
    Source:

    Here is another example, 11 Mar 2015:
    Autolog: At 3:42 "It's got lane change assist. Basically all you have to do to change lanes is just tap the indicator and the car will actually do itself when it's safe to do so."
    Source:

    Side note: Tesla Motors tweeted about this Autolog video. Lots of people watched it and now they have an incorrect idea about auto lane change. Here is that tweet:

    The Tesla blog post had incorrect information too:
    Quote: "Model S will be able to steer to stay within a lane, change lanes with the simple tap of a turn signal"
    Source: http://www.teslamotors.com/blog/dual-motor-model-s-and-autopilot

    However, the main source of the incorrect information is the Model S web page. On top menu if you click on "Model S", on that web page it says this:

    Quote: "Lane Changing: Automatically change lanes by tapping the turn signal."

    That is incorrect. The Model S can't see behind. If there is a speeding car coming from behind on the lane you want to change to, there is no way MS can see it. The ultrasonic sensor range is only 16ft (5m).

    Unfortunately in a recent interview on 17 Mar 2015, Elon repeated the incorrect information.

    Elon Musk: "Even with just that sensor suite we can make huge progress in autonomy. We can certainly make the car steer itself on a free way and do lane changes"
    Source: At 7:12

    Solution:
    1. Remove the incorrect information from the Model S page about automatic lane change.
    2. Remove the incorrect information from the blog post.
    3. Create a video that explains how lane change works and how the driver needs to check the mirrors. Put that video on this website on the Model S page. Display a code at the end of the video and ask for this code to be entered on the touchscreen before this feature is enabled.

    This misunderstanding could potentially cause accidents and bad press for Tesla. Therefore I want to bring this issue to Tesla's attention.

    Regards,
    Teo

    Screenshot from current Model S page, that has incorrect information:
    <img src=http://s11.postimg.org/5vujzc99f/image.gif&gt;

    Screenshot from Tesla blog post that has incorrect information:
    <img src=http://s1.postimg.org/98j6lxd7j/image.gif&gt;
  • edited November -1
    @EVino

    What do you think of LIDAR (dome on top) that Google are using?

    I think Elon doesn't like it due to high cost.

    <img src= "http://cdn.theatlantic.com/static/newsroom/img/mt/2014/08/googcar2/lead.jpg?na95q8&quot; width="640"/>
  • edited November -1
    Edit: cause accidents when this feature is enabled because* people
  • edited November -1
    There has been threads discussing this. You might check it out via www.volkerize.com
  • edited November -1
    @Tam, Google is in a different league. They are using lasers that can scan 360 degrees plus z axis, out to two football fields. Those systems are still expensive.
  • edited November -1
    @Teo, I believe the Model S with present sensor stack CAN do lane changes can but only in very safe parameters. The present Mobileye monocular optical system is actually better than radar at gathering data for lane keeping, and so programming it do something like "is the next lane over clear?" might be well within its data gathering parameters. Maneuvering in and out of dense traffic objects is entirely different.
  • edited November -1
    EVino,

    I don't know why you are talking about "lane keeping". I have not written anything about lane keeping. I'm talking about autopilot lane change. Please watch these short videos that demonstrate autopilot lane change.





    I have two questions:

    1. If somebody initiates an autopilot lane change without checking the mirrors, wouldn't that be dangerous because another car might be approaching fast on the lane you are changing into?

    2. The 3 Tesla sources I mentioned give the impression that you don't have to check mirrors. A misunderstanding is very likely. Those two media sources (CNET and Autoblog) have already misunderstood it as you can see in the videos. Don't you agree?
  • edited November -1
    This is what I understand:

    Tesla "autopilot" is a subset of "autonomous."

    You can hear the audio recording from OTA 6.2 announcement:

    http://my.teslamotors.com/forum/forums/horses-mouth-full-audio-elon-musk-press-conference-announce-end-range-anxiety-62

    Tesla will not enable "current suite" (that's hardware, I think) for parking-lot-to-Parking-lot autopilot due to the safety nature of unpredictable urban scenarios at midrange speed about 30 mph.

    However, the current suite is fully capable of take over your driving from merging-in-ramp to exit-ramp in highway speed (as well as low speed as in summoning your car from Supercharger parking lot to the front Hotel.)

    Thus, yes, you'll need to pay up for another "suite" for a fully autonomous system that knows how to maneuver in low speed, high speed and ALSO in mid-range speed urban scenarios.

    When? "Soon!"
  • edited November -1
    Ok but the fact remains MS will do everything S class can do if not a little more with the software update in three months. It's doing it in the simple functional way and not by adding everything you can think of. This is pretty much the fundamental difference of engineering design style between the two companies.

    Tesla is catching up in this area very rapidly and should be able to take, and widen the lead at a fast pace. It's Stuttgart vs the Silicon Valley you can pretty much see what the outcome will be.
  • edited November -1
    Teo,
    I can't speculate on that. It's easy to confuse semantics with the whole business of autonomous driving. I was merely giving a comparison of what I believe are more complete systems at present. Tesla is incremental, and their functional wording can sometimes cause semantic debates.

    Here's probably how it would work: I'm on my lane (lane #2). It will keep me on my lane for the next 40 miles. If I want to go to lane #1--it would require a visual check from me, then I would signal left. The car's sonar would perform a redundancy check that the left side is clear. The steering system would perform the maneuver, with the optical camera centering the car onto the next lane, then locking on the new lane keeping functions. Done. In the Mercedes-Benz system, they actually have rear facing radar to do the rearview check that I would need to do visually. So, in Tesla's case they would caveat that the driver needs to do this part. It's not unreasonable or impractical. We do it now manually.
  • edited November -1
    It's fairly clear the Google LIDAR system is the gold standard (and priced that way too). But it's ugly as sin.

    For fully autonomous driving, there is no reason an alternate system (cameras, radar) can't do as good a job, but that's yet to be proven. It will be exciting to see the race to come up with the sensor suite at a reasonable cost that does the job right. I do agree that the current sensor suite cannot deal with every autonomous driving issues.

    Likely lost in much of this is how utterly complex the software is to deal with any sensor suite and deal with those 1 in 10,000 cases. Often the 1 in 10,000 cases the human does something wrong too!

    I do expect there will be situations where the system will say "Sorry - Autonomous driving is unavailable - please drive carefully". For example, snow with whiteout conditions or perhaps Autobahn where lane speed differences can be 80-120 MPH.
  • edited November -1
    EVino,

    I agree with you that more hardware is needed. With a side and rear radar, autopilot could do a lot more. Elon talked about this last week. The plan is to add more hardware in a few years, maybe 3 years. Here is that video:
  • edited November -1
    I would bet on Tesla. They have sister engineers at SpaceX building sensors on the Falcon9 that would allow it to land on a moving barge.

    Regarding Google, they didn't start with the automotive mindset, so their scientists probably approached the problem without bias. I believe they are using machine learning (inductive reasoning) than if-else logic. The system has to learn and it keeps learning. Their first mistakes were probably very embarrassing, like running over animals. It will learn that mistake. Their goal is machine learning. For me, for now, I would be happy with a great adaptive cruise, some lane maneuvering around slow drivers, and collision braking, maybe one more year for Tesla's incremental push. Mercedes Benz is setting the bar for autobahn conditions.
  • edited November -1
    I like Google's approach, specifically this part:

    "Google engineers realized that asking a human passenger — who could be reading or daydreaming or even sleeping — to take over in an emergency won't work."

    Source: NYtimes article: Google’s Next Phase in Driverless Cars: No Steering Wheel or Brake Pedals
  • edited November -1
    This is a great technical discussion. My compliments.

    My main concern is indeed high speed aft approach. However, when I'm driving and checking the left rear view mirror 'objects are closer than they appear'--I'm not sure I could gauge a high approach at freeway speeds anyway. The relative speed would need high. I'm thinking that rear looking sonar might be able to do as well as I can visually. If you think about it, you only glance in that mirror for maybe 500 mS. Then forward then one last 500 mS look and initiate lane change. If someone is really cooking, you're going to get nailed.

    I'm in fact wondering if this can work at all without a left mirror aft looking radar. Would do any good in the middle of the back of the car because it could be obscured by the car following you. You wouldn't detect the oncoming vehicle until you already IN the adjacent lane where you would be impacted. So you'd probably need narrow radar (lidar?) port and starboard to handle the transition.
  • edited November -1
    Okay, missing elements in my comment should read "..would need to be high." and "Wouldn't do any good in the middle.."
    My apologies.
  • edited November -1
    @Teo-I did not misunderstand you one bit. I am clear in what you are saying. Yes, some of the media reports are wrong in saying that the car checks for safety first, wow, what else is new? TESLA HAS NEVER SAID THIS. The automatic lane change is in the context of automatic lane keeping. The car keeps in its lane, if the driver wants to change lanes, with full responsibility of clearance, the car will execute the lane change when driver hits the turn signal. THAT's all NOTHING more. YOU are reading way more into this than is being said.

    You also never answered my question if you have ever actually owned a car or if you have even ever driven one.
  • edited November -1
    @EVino: I agree with the conclusion that the MS needs additional equipment to implement autonomous driving features, which suggests newly equipped vehicles, although it may be possible to use the current wiring harness to retrofit existing autopilot models with more capable sensors than they now have to be able to implement some subset of autonomous driving features.

    But here's a mystery. Recently, I was driving my MS with autopilot at about 40 mph in a 45 mph zone. I was following what looked like a new Audi A8. I maintained about 4-6 car lengths following distance. Twice, when the Audi braked, the Tesla dash displayed a crash warning as I closed the distance to 3-4 car lengths. It has never done this before or since. TACC was not engaged. The A8 can be equipped with two rear-facing radar sensors. I'm wondering if this was the case for the Audi that I was following and, if so, was the Tesla sensor picking up radar pulses from the Audi causing it to raise the false alarm???
  • edited November -1
    Lots of Monday morning quarterbacks here. Many are accusing Musk of exaggerating. We will see. Tesla is designing the car, not the quarterbacks.
  • edited November -1
    george, could it be your forward collision warning setting set to Early?
    <img src="http://i.imgur.com/hoxNd3G.jpg&quot; /></a>

    avanti, this is a technical discussion. We have not engaged in ad hominem attacks.
  • edited November -1
    Musk said recently that in order to be autonomous the Tesla would need an upgrade of hardware sensors. But he said they will be able to do highway lane/speed management with the current sensors.

    It seems to me that Tesla needs to clarify what they mean about automatic lane changing because it could be read the wrong way. The way I understand it is that when the driver initiates a lane change the car will move from one lane to the other without hitting something that may be in the blind spot. But if a car is coming up super fast in the other lane it could get hit. And that's where the human initiating it comes in. As long as they clarify that the driver needs to check behind them first, it makes sense.
  • edited November -1
    Is the rear view camera being left out of this discussion? If car behind to close do not make lane change. If no car close behind an analysis of the new lane could be made. These are the same decisions made by a driver now.
  • edited November -1
    @avanti5010 You are absolutely right.

    @EVino <i> this is a technical discussion.

    May I ask which autopilot/self driving system you have designed so we can make some technical discussion on the system? You come on the board with less than 0.001% knowledge of how the Tesla system is designed and works and you want to do a "technical discussion"?
  • edited November -1
    Friday I drove to work as I usually do and pulled into my parking spot, approaching the sign-pole with the nose of my Model S. I was getting quite close but didn't hear the usual parking-sensor warning. Why? The nose of the car had been changed from black to white with a layer of wet, sticky snow.

    Even something as simple as a parking sensor can fail without warning. Driver assistance? Caveat Emptor.
Sign In or Register to comment.